Prognostic significance of the albumin-to-globulin ratio for advanced urothelial carcinoma treated with pembrolizumab: a multicenter retrospective study

Although the albumin-to-globulin ratio (AGR) is a promising biomarker, no study has investigated its prognostic significance for advanced urothelial carcinoma (UC). This study conformed to the REporting recommendations for tumor MARKer prognostic studies (REMARK) criteria. We retrospectively reviewed 176 patients with advanced UC treated with pembrolizumab between 2018 and 2020. We evaluated the associations between pretreatment clinicopathological variables, including the AGR and performance status (PS), with progression-free survival, cancer-specific survival, and overall survival. The Cox proportional hazards model was used for univariate and multivariable analyses. The AGR was dichotomized as < 0.95 and ≥ 0.95 based on receiver operating characteristic curve analysis. After excluding 26 cases with missing data from the total of 176 cases, 109 (73%) patients experienced disease progression, 75 (50%) died from UC, and 6 (4%) died of other causes (median survival = 12 months). Multivariate analyses identified PS ≥ 2 and pretreatment AGR < 0.95 as independent poor prognostic factors for all endpoints. Furthermore, a prognostic risk model incorporating these two variables achieved a relatively high concordance index for all endpoints. This is the first report to evaluate the significance of AGR in advanced UC. Pretreatment AGR < 0.95 may serve as a prognostic marker for advanced UC treated with pembrolizumab.


Methods
Ethical approval and informed consent. This study was approved by the Institutional Review Board Patients and study design. This retrospective study conformed to the REporting recommendations for tumor MARKer prognostic studies (REMARK) criteria 25 (Supplementary Table S1 online). We reviewed the records of 176 consecutive patients with advanced (locally-advanced or metastatic) UC treated with pembrolizumab at our seven affiliate institutions (five university hospitals and two tertiary referral hospitals) between January 2018 and July 2020. Blood tests were performed within 1 month after the initiation of pembrolizumab therapy. We excluded 26 patients whose data for AGR, NLR, or both were unavailable, leaving 150 patients for the final analysis (Fig. 1). A fixed dose of pembrolizumab (200 mg per patient) was intravenously administered every 3 weeks. All patients underwent evaluations every 1-6 months that included routine blood tests, chest X-ray, and computed tomography. The patients' charts were reviewed, and the status of each patient was assessed through office visits, telephone calls, or both.
Pretreatment AGT, NLR, and other laboratory markers. Routine pretreatment blood tests, including serum total protein and albumin levels (g/dL), neutrophil and lymphocyte counts (cells/μL), hemoglobin (g/dL), creatinine (mg/dL), and C-reactive protein (mg/dL) were performed within 1 month before the start of Statistical analysis. Receiver operating characteristic (ROC) curve analysis was used to determine the optimal cutoff values of the AGR and NLR. Sensitivity, specificity, and area under the curve (AUC) were calculated using a 2 × 2 contingency table incorporating each cutoff value. The optimal cutoff value of each ratio was determined by maximization of the Youden's index [Sensitivity − (1 − Specificity)]. The significance of the associations of the AGR with other variables were evaluated using the χ2 test or Spearman's rank correlation coefficient. Survival curves were generated using the Kaplan-Meier method and compared using log-rank tests. The Cox proportional hazard regression model was used for univariate and multivariate analyses for PFS, CSS, and OS. A prognostic model for predicting PFS, CSS, and OS was constructed according to independent prognostic factors detected using multivariate analysis. Harrell's concordance index was calculated to quantify the model's prognostic discrimination 28 . All statistical analyses, except the concordance index, were performed using JMP Pro version 14.0.0 (SAS Institute, Cary, NC, USA). Harrell's concordance index was calculated by a biostatistician (Y.U.) using SAS version 9.4. P < 0.05 indicates a significant difference.
The χ 2 test revealed that the pathological Eastern Cooperative Oncology Group Performance Status (ECOG PS, ≥ 2, P < 0.01) and bone metastasis (yes, P < 0.01) were significantly associated with AGR < 0.95. In contrast, the other variables (sex, primary site, resection of primary site, lymph node metastasis, lung metastasis, liver metastasis, and number of prior regimens) were not. Spearman's rank correlation coefficient showed a strongly significant negative correlation between C-reactive protein and the AGR (ρ = − 0.71, P < 0.01), as well as a weakly significant negative correlation between the NLR and the AGR (ρ = − 0.41, P < 0.01) and a weakly significant positive correlation between hemoglobin and the AGR (ρ = + 0.46, P < 0.01).
Kaplan-Meier curves with log-rank tests showed significant associations of AGR < 0.95 and NLR ≥ 3 with shorter PFS, CSS, and OS (Fig. 2). Multivariate Cox proportional hazard regression analyses identified pretreatment AGR < 0.95 as an independent indicator of poor prognosis for PFS together with the following: ECOG PS ≥ 2 and liver metastasis ( Table 2); CSS with ECOG PS ≥ 2 (Table 3); and OS with ECOG PS ≥ 2 (Table 4). Pretreatment NLR ≥ 3 showed a non-significant trend for shorter CSS, whereas liver metastasis did for shorter OS (both P = 0.08) (Tables 3 and 4).
A prognostic risk model designed to predict PFS, CSS, and OS was developed according to the two shared risk factors in the multivariate analyses for all endpoints as follows: no risk, AGR ≥ 0.95 and PS ≤ 1; one risk, AGR < 0.95 or PS ≥ 2; and two risks, AGR < 0.95 and PS ≥ 2. A significant difference was found among the survival profiles of the three risk groups (Fig. 3). The Harrell's concordance indices of this model were PFS, 0.63; CSS, 0.68; and OS, 0.67.

Discussion
To our knowledge, this study is the first study to assess the prognostic significance of the AGR in patients with advanced UC. Here we analyzed a multi-institutional cohort of 150 patients with advanced UC patients treated with pembrolizumab. We found that pretreatment AGR < 0.95 and ECOG PS ≥ 2 were independent predictors of PFS, CSS, and OS. Furthermore, we developed a prognostic risk model incorporating these two variables, which classified patients into three risk groups with significantly different PFS, CSS, and OS values. The model achieved a relatively high Harrell's concordance index for all study endpoints.
Although the concept of AGR was introduced long ago, it was not applied to oncology until the 2010s 14,15 . Several studies assessed the prognostic significance of the AGR of patients with UC [16][17][18][19][20][21][22][23][24] in settings of bladder cancer treated with radical cystectomy 16,17 , non-muscle-invasive bladder cancer 18 , and upper tract UC treated with radical nephroureterectomy [19][20][21][22][23][24] . Most of these studies [16][17][18][19][20][21][22][23] demonstrate the utility of the AGR as a readily available predictive biomarker for patients with UC. To our knowledge, the present study is the first to demonstrate the potential utility of the AGR in the setting of advanced UC, and thus adds further evidence in this field.
The link between a low AGR and poor outcomes of patients with cancer is not fully established but may be explained in general terms as follows 23,29 : First, previous studies show that poor nutritional status or hypoalbuminemia is a predictor of poor prognosis for patients with cancer 29 . Second, chronic inflammation involving serum globulins is required for tumor proliferation, immune evasion, and metastasis. Evidence indicates that serum globulins secreted by tumor-related cells promote tumor development, immunosuppression, and metastasis 29 . Third, a low AGR may thus more sensitively reflect the degree of poor nutritional status (hypoalbuminemia) and tumor progression (hyperglobulinemia) than either measure alone, and may therefore serve as a highly significant prognostic biomarker 23 .
Similar to the concept of the AGR (i.e. use of a ratio), the NLR was developed in 2001 by Roman Zahorec 7 and has subsequently been investigated in oncology. Numerous studies show an association with an increased NLR with worse outcomes of certain malignancies including UC [8][9][10][11][12][13] . The NLR sensitively reflects the degree of tumor progression; this is because both an increased neutrophil-dependent inflammatory reaction and a decreased lymphocyte-mediated anti-tumor immune response contribute to the elevation of the NLR 8,30 .
We previously reported the prognostic significance of the NLR using a multi-institutional cohort of 185 patients with advanced UC undergoing first-line chemotherapy 8 . Pretreatment NLR ≥ 3 was identified as an independent predictor of CSS and OS together with ECOG PS ≥ 2 and liver metastasis, whereas the AGR was not evaluated 8 . In the present study, pretreatment NLR ≥ 3 was significantly associated with all endpoints of PFS, CSS, and OS on univariate analyses (Tables 2, 3, 4), and showed a non-significant trend for shorter CSS on a multivariate analysis (P = 0.08) ( Table 3). These data indicate that the NLR may still serve as a valid biomarker in the setting of later-line pembrolizumab.  www.nature.com/scientificreports/ We further reported the critical impact of liver metastasis on worse outcomes in the said study 8 . In the present study, liver metastasis was significantly associated with all endpoints of PFS, CSS, and OS on univariate analyses (Tables 2, 3, 4), and was identified as an independent predictor of shorter PFS (Table 2) with showing a nonsignificant trend for shorter OS (P = 0.08) ( Table 4) on multivariate analyses. Although liver metastasis was not incorporated into the final risk model applied here, it undoubtedly serves as an essential prognostic marker for advanced UC, even in the era of immune checkpoint inhibitors.
The limitations of this study include its retrospective design and the limited number of patients. Further studies with larger populations are required to validate our results.
In conclusion, pretreatment AGR < 0.95 may serve as a prognostic marker for patients with advanced UC treated with pembrolizumab. Our newly developed prognostic risk model, including pretreatment AGR and ECOG PS, may serve as an excellent discriminator of survival. Table 2. Univariate and multivariate Cox proportional hazard regression analyses of PFS. AGR albumin-toglobulin ratio, CI confidence interval, ECOG PS Eastern Cooperative Oncology Group Performance Status, eGFR estimated glomerular filtration rate, HR hazard ratio, LLN lower limit of normal, NLR neutrophil-tolymphocyte ratio, PFS progression-free survival. † LLN of hemoglobin was set at 13.0 g/dL for men and 11.5 g/ dL for women. *Statistically significant.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request. www.nature.com/scientificreports/ the study, helped to draft the manuscript and were involved in revising it critically for important intellectual content. All authors read and approved the final manuscript.